Evaluating and Mitigating Discrimination in Language Model Decisions